Joe shares three big challenges he’s seen in AI labs and how organizations are tackling them.
Hey, welcome to #TheGouge with Ponch and Cujo. I’m Ponch and today I’m here with my friend Joe Justice and we’re talking about building AI labs. Right Joe? What’s the gouge on that?
In the last year, in the classes we teach, people are saying: How do we boot up our AI lab faster? We’re seeing this in companies all around the world. They want more ML, they want more natural language processing, they want more AI. And, the challenge they’re finding, when they call us in to help boot up the lab, is these people have an academic model which is not meeting the speed demands they want in the company. They’re saying, 1, How do we prioritize, so it’s not all these AI researchers making their own project, but how do we have something where all the projects combine and make a whole lot of money for the company? And, how do we keep these people’s interest? Because the AI lab right down the street is going to pay them just as much, if not more, all the time. It’s a very hot market right now. So, talent retention is a top priority issue, focus of that talent is a top priority issue, and monetization of the product is a top priority issue.
For those reasons, they’re saying, well let’s implement a rigorous product owner to prioritize what we’re going to do and most of their job isn’t “let me tell you what’s important,” it’s figuring out what’s important, and then making it so fun that most of these researchers opt in to do it. Having a big visible backlog and rotating who plays the product owner role seems to be solving that right now. So many of these labs that I’ve had a chance to play in the last year, each Sprint they’ll rotate who plays the product owner role and they’ll have a pair of whoever was last to try to keep consistency of prioritization. But that creates the buy-in, so these people actually work on the one top priority. Then, the tools they’re working on, they arise very dynamically right now, on the daily. So, the daily standup is often a tools discussion as well. Are you using TensorFlow, are you using pie torch, why? They also have a starter sandbox set for people coming in, so someone right out of academia has this starter set, but very quickly they’re pulling tools within hours, because it’s that dynamic a space right now.
Then the monetization is turning out to be a lot easier than I think everybody thought it would, once they have a prioritization structure. It looked like the difficulty of monetizing these AI labs wasn’t that they didn’t have monetizable solutions, it’s that they weren’t able to get people to actually work on the top priority. But by rotating who plays it, gamifying it – I’ve seen people make it look just like a board game in the lab, that seems to speak to their structure right now – they get the buy-in. That putting the rewards, that also speak to the academic model but also speak to the straight business model, right in the backlog seems to be working out, too. So, writing research papers are backlog items every Sprint. Every Sprint, everyone gets their name on another research paper. Some companies view patents as a liability, some as an advantage. If they view patents as an advantage, they put it right in the backlog. And they do it every Sprint. That’s what I’m learning now.
Awesome. And that, my friends, is the gouge on AI labs. However, I have no idea what a pie torch is.